神经隐式表征是一种新兴的形状表示范式,但多数传统隐式表示方法如DeepSDF等未考虑整体形状的局部特征信息,存在拓扑细节精度不足的问题。为解决上述问题,提出了一种由部件隐向量驱动的隐式三维重建方法,即构建部件的隐式场以最小化模...神经隐式表征是一种新兴的形状表示范式,但多数传统隐式表示方法如DeepSDF等未考虑整体形状的局部特征信息,存在拓扑细节精度不足的问题。为解决上述问题,提出了一种由部件隐向量驱动的隐式三维重建方法,即构建部件的隐式场以最小化模型预测的整体形状目标点有符号距离值LGI-RIF(Reconstruction of Implicit Fields with Local and Global Integration),能从观测数据中重建几何形状。该方法在一个低维的潜在编码空间中训练神经网络并在解码器框架中联合优化,设计EFP、EFCS和R3DS这3个模块,在EFP模块中由设计的变分自编码器学习部件的特征向量分布,在EFCS模块中构建自动解码器学习整体形状的SDF隐式表达,在R3DS模块中重建整体三维形状。实验结果表明:LGI-RIF在ShapeNet和ModelNet 10数据集上的重建精度得到了进一步提升,在可视化重构中具有良好的视觉效果。展开更多
We propose a multi-field implicit finite element method for analyzing the electromechanical behavior of dielectric elastomers. This method is based on a four-field variational principle, which includes displacement an...We propose a multi-field implicit finite element method for analyzing the electromechanical behavior of dielectric elastomers. This method is based on a four-field variational principle, which includes displacement and electric potential for the electromechanical coupling analysis, and additional independent fields to address the incompressible constraint of the hyperelastic material. Linearization of the variational form and finite element discretization are adopted for the numerical implementation. A general FEM program framework is devel- oped using C＋＋ based on the open-source finite element library deal.II to implement this proposed algorithm. Numerical examples demonstrate the accuracy, convergence properties, mesh-independence properties, and scalability of this method. We also use the method for eigenvalue analysis of a dielectric elastomer actuator subject to electromechanical loadings. Our finite element implementation is available as an online supplementary material.展开更多
文摘神经隐式表征是一种新兴的形状表示范式,但多数传统隐式表示方法如DeepSDF等未考虑整体形状的局部特征信息,存在拓扑细节精度不足的问题。为解决上述问题,提出了一种由部件隐向量驱动的隐式三维重建方法,即构建部件的隐式场以最小化模型预测的整体形状目标点有符号距离值LGI-RIF(Reconstruction of Implicit Fields with Local and Global Integration),能从观测数据中重建几何形状。该方法在一个低维的潜在编码空间中训练神经网络并在解码器框架中联合优化,设计EFP、EFCS和R3DS这3个模块,在EFP模块中由设计的变分自编码器学习部件的特征向量分布,在EFCS模块中构建自动解码器学习整体形状的SDF隐式表达,在R3DS模块中重建整体三维形状。实验结果表明:LGI-RIF在ShapeNet和ModelNet 10数据集上的重建精度得到了进一步提升,在可视化重构中具有良好的视觉效果。
基金the support under A*STAR SERC grant (132-183-0025)
文摘We propose a multi-field implicit finite element method for analyzing the electromechanical behavior of dielectric elastomers. This method is based on a four-field variational principle, which includes displacement and electric potential for the electromechanical coupling analysis, and additional independent fields to address the incompressible constraint of the hyperelastic material. Linearization of the variational form and finite element discretization are adopted for the numerical implementation. A general FEM program framework is devel- oped using C＋＋ based on the open-source finite element library deal.II to implement this proposed algorithm. Numerical examples demonstrate the accuracy, convergence properties, mesh-independence properties, and scalability of this method. We also use the method for eigenvalue analysis of a dielectric elastomer actuator subject to electromechanical loadings. Our finite element implementation is available as an online supplementary material.
